Machine learning approaches present significant opportunities for optimizing existing machines and production systems. Particularly in hot rolling processes, great potential for optimization can be exploited. Radial-axial ring rolling is a crucial process utilized to manufacture seamless rings. However, the failure of the mandrel represents a defect within the ring rolling process that currently cannot be adequately explained. Mandrel failure is unpredictable, occurs without a directly identifiable reason, and can appear several times a week depending on the ring rolling mill and capacity utilization. Broken rolls lead to unscheduled production downtimes, defective rings and can damage other machine parts. Considering the extensive recording of production data in ring rolling, the implementation of machine learning models for the prediction of such roll breaks offers great potential. To present a comprehensive overview of the potential influencing factors which are possibly relevant to the lifetime of mandrels, a systematic literature review (SLR) focusing on work roll wear in hot rolling processes is conducted. Based on the results of the SLR, a first selection of features and the used investigation procedures are presented. The insights can be used for the prediction of mandrel failure with machine learning models in further work.
|